hf-l2-skill-router

Files (7)

• SKILL.md
• .gitkeep · evals/
• evals.json · evals/
• .gitkeep · metrics/
• rich-gate-scorecard.md · metrics/
• .gitkeep · references/
• routing-map.md · references/

SKILL.md (first lines)

# The Iron Law

```
Map every meta-builder task domain to the correct hf-* skill bundle. Max 3 skills per bundle. Priority ordering is binding. Cross-route to hm-* only when no hf-* skill covers the domain.
```

# HF Skill Router

## Overview

This router maps meta-builder task domains to hf-* skill loading bundles. It is used by hf-orchestrator and hf-coordinator to dispatch skills before meta-builder task execution. It does NOT execute skills — it only determines which skills to load and in what priority order.

This is a **dispatch router** specifically for the meta-builder lineage. While hm-l2-skill-router handles product-dev domains, hf-l2-skill-router handles meta-concept creation, auditing, and assembly domains.

**FLEXIBLE lineage note:** This router may route to hm-* skills for cross-validation when no hf-* skill covers the domain. The FLEXIBLE designation means hm-* skills are allowed as secondary/fill-in bundles, not primary.

**Eight task domains, eight skill bundles:**

| Domain | Skills | Max |
|--------|--------|-----|
| Agent Building | hf-l2-agents-and-subagents-dev + hf-l2-agent-composition | 2 |
| Skill Authoring | hf-l2-use-authoring-skills + hf-l2-skill-synthesis | 2 |
| Command Dev | hf-l2-command-dev + hf-l2-command-parser | 2 |
| Tool Building | hf-l2-custom-tools-dev | 1 |
| Audit | hf-l2-use-authoring-skills + hf-l2-agents-md-sync + gate-l3-evidence-truth | 3 |
| Refactor | hf-l2-agents-md-sync + hm-l2-refactor | 2 |
| Synthesis | hf-l2-skill-synthesis + hm-l3-synthesis | 2 |
| Delegation | hf-l2-delegation-gates | 1 |

## On Load

1. Read `references/routing-map.md` — the complete dispatch map with domain signal tables and loading rules
2. Identify the task domain from the delegation context
3. Classify the task into one of the eight domains
4. Return the skill bundle with priority ordering
5. If cross-routing to hm-* skills, note the FLEXIBLE lineage justification

## Trigger Phrases

- "load hf skills for task" / "which hf skills do I need"
- "map meta-builder domain to skills" / "hf skill loading bundle"
- "meta-builder skill router" / "hf skill dispatch"
- "route meta-builder task" / "determine hf skill bundle"
- "hf skill selection" / "meta-builder skill dispatch map"
- "task-to-hf-skill mapping" / "hf routing table"
- "load skills for meta builder" / "which skills for creating agents"
- "meta-builder dispatch" / "load hf bundle"

## Domain Dispatch Map

### Domain 1: Agent Building

**Intent signals:** "create an agent", "add an agent", "write a subagent", "agent frontmatter", "define agent permissions", "agent composition"

| Priority | Skill | Role |
|----------|-------|------|
| 1 | `hf-l2-agents-and-subagents-dev` | Agent architecture, permissions, subagent dispatch protocols |
| 2 | `hf-l2-agent-composition` | XML markup grammar, step protocols, structured return formats |

**Loading order:** hf-l2-agents-and-subagents-dev (architecture + permissions) → hf-l2-agent-composition (composition format)

### Domain 2: Skill Authoring

**Intent signals:** "create a skill", "write a SKILL.md", "audit this skill", "skill frontmatter", "skill package", "synthesize a skill"

| Priority | Skill | Role |
|----------|-------|------|
| 1 | `hf-l2-use-authoring-skills` | Skill quality framework, TDD for skills, cross-platform compatibility |
| 2 | `hf-l2-skill-synthesis` | Pattern extraction from repos, eval frameworks, skill scaffolds |

**Loading order:** hf-l2-use-authoring-skills (quality framework) → hf-l2-skill-synthesis (pattern extraction)

### Domain 3: Command Dev

**Intent signals:** "create a command", "add a command", "write a custom command", "$ARGUMENTS parsing", "command with bash injection"

| Priority | Skill | Role |
|----------|-------|------|
| 1 | `hf-l2-command-dev` | Command structure, YAML frontmatter, $ARGUMENTS, !bash, non-interactive shell safety |
| 2 | `hf-l2-command-parser` | Propositional command parsing, named arguments, flag extraction |

**Loading order:** hf-l2-command-dev (structure + safety) → hf-l2-command-parser (argument parsing)

### Domain 4: Tool Building

**Intent signals:** "create a custom tool", "build an OpenCode plugin", "write a tool with Zod", "add a plugin hook"

| Priority | Skill | Role |
|----------|-------|------|
| 1 | `hf-l2-custom-tools-dev` | Tool definitions with Zod schemas, plugin lifecycle hooks, CLI scripts |

**Note:** Single-skill bundle. Tool building is a focused domain.

### Domain 5: Audit

**Intent signals:** "audit skills", "audit agents", "audit commands", "drift detection", "quality compliance", "structural integrity", "audit project"

| Priority | Skill | Role |
|----------|-------|------|
| 1 | `hf-l2-use-authoring-skills` | Cross-primitive quality scoring |
| 2 | `hf-l2-agents-md-sync` | Drift detection between AGENTS.md and actual state |
| 3 | `gate-l3-evidence-truth` | Terminal evidence gate — validates audit findings have runtime proof |

**Loading order:** hf-l2-use-authoring-skills (quality check) → hf-l2-agents-md-sync (drift check) → gate-l3-evidence-truth (evidence validation)

### Domain 6: Refactor

**Intent signals:** "refactor skills", "refactor agents", "clean up meta-concepts", "restructure skill", "reduce technical debt in skills"

| Priority | Skill | Role |
|----------|-------|------|
| 1 | `hf-l2-agents-md-sync` | Sync pattern for meta-concept refactoring |
| 2 | `hm-l2-refactor` | Surgical vs structural refactoring decisions (FLEXIBLE cross-route) |

**Loading order:** hf-l2-agents-md-sync (sync scope) → hm-l2-refactor (refactoring methodology)

**FLEXIBLE note:** hm-l2-refactor is a product-dev skill. Cross-routed because no hf-* skill covers structural refactoring methodology. This is a documented, intentional cross-lineage route.

### Domain 7: Synthesis

From shynlee04/hivemind-plugin, MIT.

More from shynlee04/hivemind-plugin